Um estudo comparativo entre redes neurais e séries temporais para predições de valores de ações no mercado de fintechs

Abstract

Predicting stock prices in the financial market is one of the most challenging tasks in machine learning due to the dynamic, nonlinear, and complex nature of these markets. Fintechs, companies that integrate technology and financial services, form a growing sector in the market, standing out for their innovation, scalability, and potential for high returns. However, they are also accompanied by high investment risks, which justifies the importance of carefully analyzing the market and these companies’ growth strategies before investing. This research presents a comparative study of four predictive methods, namely moving averages, moving averages with exponential smoothing, ARIMA (Autoregressive Integrated Moving Average), and LMST neural networks, for predicting stock prices of Brazilian fintech companies listed on the B3 stock exchange. The study assessed the accuracy of stock price predictions for the following day using Root Mean Square Error (RMSE) as a performance metric. The results demonstrate that the ARIMA method outperforms the others in all scenarios, regardless of the number of previous days considered in the analysis, effectively capturing complex market patterns and assisting investors in decision-making